Add dir_fd to os.path.lexists() & os.path.isdir()

[nineteendo]

Feature or enhancement

Proposal:

glob currently needs its own implementation of os.path.lexists() & os.path.isdir() to support dir_fd:

cpython/Lib/glob.py

Lines 201 to 222 in f74e512

	def _lexists(pathname, dir_fd):
	# Same as os.path.lexists(), but with dir_fd
	if dir_fd is None:
	return os.path.lexists(pathname)
	try:
	os.lstat(pathname, dir_fd=dir_fd)
	except (OSError, ValueError):
	return False
	else:
	return True
	def _isdir(pathname, dir_fd):
	# Same as os.path.isdir(), but with dir_fd
	if dir_fd is None:
	return os.path.isdir(pathname)
	try:
	st = os.stat(pathname, dir_fd=dir_fd)
	except (OSError, ValueError):
	return False
	else:
	return stat.S_ISDIR(st.st_mode)

We could refactor this by adding dir_fd to os.path.lexists() & os.path.isdir():

-def lexists(path):
+def lexists(path, *, dir_fd: int | None = None):
     """Test whether a path exists.  Returns True for broken symbolic links"""
     try:
-        os.lstat(path)
+        os.lstat(path, dir_fd=dir_fd)
     except (OSError, ValueError):
         return False
     return True

-def isdir(s):
+def isdir(s, *, dir_fd: int | None = None):
     """Return true if the pathname refers to an existing directory."""
     try:
-        st = os.stat(s)
+        st = os.stat(s, dir_fd=dir_fd)
     except (OSError, ValueError):
         return False
     return stat.S_ISDIR(st.st_mode)

Note: nt._path_isdir() (& nt._path_lexists() when #117842 lands) need to raise an error for this.

Has this already been discussed elsewhere?

This is a minor feature, which does not need previous discussion elsewhere

Links to previous discussion of this feature:

No response

[nineteendo]

cc @serhiy-storchaka before I start implementing this: is this something you would support?

[serhiy-storchaka]

It is trivially implemented via os.stat. glob implements private helpers for historical reasons -- they should behave exactly like os.path versions.

If there will be other uses of such functions in several different places, it will be worth to add this feature.

[nineteendo]

It is trivially implemented via os.stat.

Note that that's overkill on Windows. You're right because dir_fd isn't supported on Windows, sorry.

glob implements private helpers for historical reasons

They could still be simplified in that case:

def _lexists(pathname, dir_fd):
    # Same as os.path.lexists()
    return os.path.lexists(pathname, dir_fd=dir_fd)

def _isdir(pathname, dir_fd):
    # Same as os.path.isdir()
    return os.path.isdir(pathname, dir_fd=dir_fd)

[nineteendo]

@barneygale, do you think this can speed things up? No, it can't, but it could also be used for #117737.

[eryksun]

Note: nt._path_isdir() (& nt._path_lexists() when #117842 lands) would also need to be updated.

The builtin _path_* functions on Windows would have to be modified to raise NotImplementedError if the new dir_fd argument isn't None.

FYI, the NT kernel supports opening relative to a handle, which the Windows API makes use of when opening relative to the working directory. However, it's not directly exposed in the Windows API. A fundamental problem is that ".." components are resolved logically on Windows by the user-mode runtime library. Thus using a safe open on NT requires that the normalized relative path doesn't begin with a ".." component.

[barneygale]

I don't feel strongly about it. I note it would be the first os.path function to accept dir_fd.

[nineteendo]

The builtin _path_* functions on Windows would have to be modified to raise NotImplementedError if the new dir_fd argument isn't None.

I see, I assumed Windows supported it too. Then it can only be used for refactoring glob.

I note it would be the first os.path function to accept dir_fd.

At the moment it only seems to be needed for glob, so I'll leave it up to serhiy to decide.

[nineteendo]

I'm closing this as there's not much support. Feel free to re-open when you change your mind.